The continuing saga of the hyperbolic tangent

In article , John Popelish wrote: [...]

The earth seriously rolls off any frequencies above 1KHz. If you apply a high frequency boost to the analog signal before the digitizer and a low pass in software land, you can easily end up with more bits in the result than the bits in the converter.

It gives you a dithering effect for free.

I'll see your cheap-ass AD7791, ADS1271 converters, and raise you an industry-standard high-performance AKM ak5394A or Cirrus Logic cs5381 multi-bit delta-sigma 192kHz 24-bit two-channel converter, the latter only $18.80 single-pieces at Newark. Or, if you insist on cheap, cheap, go with the remnants of the older Crystal cs5360 24-bit 48kHz 2-channel converters, in stock at Newark, $8.68 each. Or the tiny 8-pin 24-bit cs5333-BZ, $8.91. If cost is foremost, try cs5340 24-bit "karaoke A/D" - factory-priced at $1.92 qty 10k.

These little high-performance A/Ds are amazing. For example, if you want to make a small super-simple one-channel machine, try the 8-pin cs5511, $4.36 ea at Newark, $2.93 qty 1k (Cirrus Logic has the datasheet). This fellow consumes a miserly 1.9mW and spews a 20-bit data stream at clock/612, e.g., 200Hz with a 122kHz clock. Sheesh, forget 10-bit converters and an LM13700. :>)

Your hardware project makes little sense to me, much as I liked the good old OPAs, Gilbert-cell variants, etc. Now, it would be interesting to take a clever high-performance 3-axis sensor with signal conditioning plus cheap 3-channel 24-bit converters, and add a digital tanh signal post-processing step to reduce the raw long-term data-archiving requirements. 200 3-axis Sa/s = 18.6GSa/yr = 75GB/yr = 750GB/decade of 32-bit words, not including time stamping and engineering data. You could find clever algorithms to reduce that burden significantly. And to analyze and display data.

I have a suggestion that'll simplify digitizing the data, and give you frequency response down to DC in the process: Replace the pickup coil with differential capacitance plates and use an Analog Devices AD7746 24-bit capacitance-to-digital converter.

Check out Analog Devices' differential capacitance approach, you'll find it can wipe the pants of a differential photometer.

Its worse than that. One of my neighbors is a black smith, and has just installed a big pneumatic hammer. I expect to see that signature quite often. I am not using a geophone, but am building my own design of a Lehman (garden gate) pendulum.

I will have to characterize its electrical and mechanical frequency response, but I expect to go easily down to 0.1 Hz. That is where the energy is for quakes arriving from the far side of the planet. I intend to come up with a fairly constant delay bandpass design (that includes the mechanical poles caused by pendulum resonant frequency) that covers the range of something like 0.1 to 10 Hz. When I get into testing the filter design, I may post some results for comment and suggestion.

I expect to eventually add a differential photometer output to the pendulum to make it a tilt meter. I want to operate both sensors on the same mechanism to compare their noise performance.

In article , John Popelish wrote: [...]

If you are near a road, beware that trucks make frequencies above your

If your geophone's characteristics are well damped and well known, you should be able to correct the data to push the low end cut off down by about a decade. A electromagnetic damping is usually built into a modern geophone.

Turns out to be pretty simple with a PTC thermistor, see....

Newsgroups: alt.binaries.schematics.electronic Subject: TANH Curve Generation - TANH-Compressor.pdf Message-ID:

...Jim Thompson

(snip)

Thanks. I'll take a look. The only charm of the photometer approach is that an LED, a couple photo diodes and opamps are all that is needed.

In case anyone is interested in this continuing exploration of arbitrary function generation, I discovered, today, that I get about half as much ripple in the hyperbolic gain function that I am trying to generate (equal % change in signal causes equal increment in output) over the same range, if I cascade the two tanh generators instead of paralleling them.

I am scratching my head to think of other ways to combine these two tanh functions that might work even better.